kaniko is a tool to build container images from a Dockerfile, inside a container or Kubernetes cluster.
kaniko doesn't depend on a Docker daemon and executes each command within a Dockerfile completely in userspace. This enables building container images in environments that can't easily or securely run a Docker daemon, such as a standard Kubernetes cluster.
kaniko is meant to be run as an image, gcr.io/kaniko-project/executor.
We do not recommend running the kaniko executor binary in another image, as it might not work.
If you are interested in contributing to kaniko, see DEVELOPMENT.md and CONTRIBUTING.md.
Table of Contents generated with DocToc
- How does kaniko work?
- Known Issues
- Demo
- Using kaniko
- Security
- Comparison with Other Tools
- Community
- Limitations
The kaniko executor image is responsible for building an image from a Dockerfile and pushing it to a registry. Within the executor image, we extract the filesystem of the base image (the FROM image in the Dockerfile). We then execute the commands in the Dockerfile, snapshotting the filesystem in userspace after each one. After each command, we append a layer of changed files to the base image (if there are any) and update image metadata.
kaniko does not support building Windows containers.
To use kaniko to build and push an image for you, you will need:
- A build context, aka something to build
- A running instance of kaniko
kaniko's build context is very similar to the build context you would send your Docker daemon for an image build; it represents a directory containing a Dockerfile which kaniko will use to build your image.
For example, a COPY command in your Dockerfile should refer to a file in the build context.
You will need to store your build context in a place that kaniko can access. Right now, kaniko supports these storage solutions:
- GCS Bucket
- S3 Bucket
- Local Directory
Note: the local directory option refers to a directory within the kaniko container. If you wish to use this option, you will need to mount in your build context into the container as a directory.
If using a GCS or S3 bucket, you will first need to create a compressed tar of your build context and upload it to your bucket. Once running, kaniko will then download and unpack the compressed tar of the build context before starting the image build.
To create a compressed tar, you can run:
tar -C <path to build context> -zcvf context.tar.gz .Then, copy over the compressed tar into your bucket. For example, we can copy over the compressed tar to a GCS bucket with gsutil:
gsutil cp context.tar.gz gs://<bucket name>When running kaniko, use the --context flag with the appropriate prefix to specify the location of your build context:
| Source | Prefix |
|---|---|
| Local Directory | dir://[path to a directory in the kaniko container] |
| GCS Bucket | gs://[bucket name]/[path to .tar.gz] |
| S3 Bucket | s3://[bucket name]/[path to .tar.gz] |
If you don't specify a prefix, kaniko will assume a local directory.
For example, to use a GCS bucket called kaniko-bucket, you would pass in --context=gs://kaniko-bucket/path/to/context.tar.gz.
There are several different ways to deploy and run kaniko:
Requirements:
- Standard Kubernetes cluster (e.g. using GKE)
- Kubernetes Secret
- A build context
To run kaniko in a Kubernetes cluster, you will need a standard running Kubernetes cluster and a Kubernetes secret, which contains the auth required to push the final image.
To create a secret to authenticate to Google Cloud Registry, follow these steps:
- Create a service account in the Google Cloud Console project you want to push the final image to with
Storage Adminpermissions. - Download a JSON key for this service account
- Rename the key to
kaniko-secret.json - To create the secret, run:
kubectl create secret generic kaniko-secret --from-file=<path to kaniko-secret.json>Note: If using a GCS bucket in the same GCP project as a build context, this service account should now also have permissions to read from that bucket.
The Kubernetes Pod spec should look similar to this, with the args parameters filled in:
apiVersion: v1
kind: Pod
metadata:
name: kaniko
spec:
containers:
- name: kaniko
image: gcr.io/kaniko-project/executor:latest
args: ["--dockerfile=<path to Dockerfile within the build context>",
"--context=gs://<GCS bucket>/<path to .tar.gz>",
"--destination=<gcr.io/$PROJECT/$IMAGE:$TAG>"]
volumeMounts:
- name: kaniko-secret
mountPath: /secret
env:
- name: GOOGLE_APPLICATION_CREDENTIALS
value: /secret/kaniko-secret.json
restartPolicy: Never
volumes:
- name: kaniko-secret
secret:
secretName: kaniko-secretThis example pulls the build context from a GCS bucket. To use a local directory build context, you could consider using configMaps to mount in small build contexts.
Running kaniko in gVisor provides an additional security boundary.
You will need to add the --force flag to run kaniko in gVisor, since currently there isn't a way to determine whether or not a container is running in gVisor.
docker run --runtime=runsc -v $(pwd):/workspace -v ~/.config:/root/.config \
gcr.io/kaniko-project/executor:latest \
--dockerfile=<path to Dockerfile> --context=/workspace \
--destination=gcr.io/my-repo/my-image --forceWe pass in --runtime=runsc to use gVisor.
This example mounts the current directory to /workspace for the build context and the ~/.config directory for GCR credentials.
Requirements:
To run kaniko in GCB, add it to your build config as a build step:
steps:
- name: gcr.io/kaniko-project/executor:latest
args: ["--dockerfile=<path to Dockerfile within the build context>",
"--context=dir://<path to build context>",
"--destination=<gcr.io/$PROJECT/$IMAGE:$TAG>"]kaniko will build and push the final image in this build step.
Requirements:
We can run the kaniko executor image locally in a Docker daemon to build and push an image from a Dockerfile.
- Load the executor image into the Docker daemon by running:
make images- Run kaniko in Docker using
run_in_docker.sh:
./run_in_docker.sh <path to Dockerfile> <path to build context> <destination of final image>kaniko currently can cache layers created by RUN commands in a remote repository.
Before executing a command, kaniko checks the cache for the layer.
If it exists, kaniko will pull and extract the cached layer instead of executing the command.
If not, kaniko will execute the command and then push the newly created layer to the cache.
Users can opt in to caching by setting the --cache=true flag.
A remote repository for storing cached layers can be provided via the --cache-repo flag.
If this flag isn't provided, a cached repo will be inferred from the --destination provided.
kaniko can cache images in a local directory that can be volume mounted into the kaniko image.
To do so, the cache must first be populated, as it is read-only. We provide a kaniko cache warming
image at gcr.io/kaniko-project/warmer:
docker run -v $(pwd):/workspace gcr.io/kaniko-project/warmer:latest --cache-dir=/workspace/cache --image=<image to cache> --image=<another image to cache>--image can be specified for any number of desired images.
This command will cache those images by digest in a local directory named cache.
Once the cache is populated, caching is opted into with the same --cache=true flag as above.
The location of the local cache is provided via the --cache-dir flag, defaulting at /cache as with the cache warmer.
See the examples directory for how to use with kubernetes clusters and persistent cache volumes.
kaniko uses Docker credential helpers to push images to a registry.
kaniko comes with support for GCR and Amazon ECR, but configuring another credential helper should allow pushing to a different registry.
The Amazon ECR credential helper is built in to the kaniko executor image. To configure credentials, you will need to do the following:
- Update the
credHelperssection of config.json with the specific URI of your ECR registry:
{
"credHelpers": {
"aws_account_id.dkr.ecr.region.amazonaws.com": "ecr-login"
}
}You can mount in the new config as a configMap:
kubectl create configmap docker-config --from-file=<path to config.json>- Create a Kubernetes secret for your
~/.aws/credentialsfile so that credentials can be accessed within the cluster.
To create the secret, run:
kubectl create secret generic aws-secret --from-file=<path to .aws/credentials>The Kubernetes Pod spec should look similar to this, with the args parameters filled in:
apiVersion: v1
kind: Pod
metadata:
name: kaniko
spec:
containers:
- name: kaniko
image: gcr.io/kaniko-project/executor:latest
args: ["--dockerfile=<path to Dockerfile within the build context>",
"--context=s3://<bucket name>/<path to .tar.gz>",
"--destination=<aws_account_id.dkr.ecr.region.amazonaws.com/my-repository:my-tag>"]
volumeMounts:
- name: aws-secret
mountPath: /root/.aws/
- name: docker-config
mountPath: /kaniko/.docker/
restartPolicy: Never
volumes:
- name: aws-secret
secret:
secretName: aws-secret
- name: docker-config
configMap:
name: docker-configThis flag allows you to pass in ARG values at build time, similarly to Docker. You can set it multiple times for multiple arguments.
Set this flag as --cache=true to opt in to caching with kaniko.
Set this flag to specify a local directory cache for base images. Defaults to /cache.
This flag must be used in conjunction with the --cache=true flag.
Set this flag to specify a remote repository which will be used to store cached layers.
If this flag is not provided, a cache repo will be inferred from the --destination flag.
If --destination=gcr.io/kaniko-project/test, then cached layers will be stored in gcr.io/kaniko-project/test/cache.
This flag must be used in conjunction with the --cache=true flag.
Set this flag to clean the filesystem at the end of the build.
Set this flag if you want to pull images from a plain HTTP registry. It is supposed to be used for testing purposes only and should not be used in production!
Set this flag if you only want to build the image, without pushing to a registry.
Set this flag to strip timestamps out of the built image and make it reproducible.
This flag takes a single snapshot of the filesystem at the end of the build, so only one layer will be appended to the base image.
Set this flag to skip TLS certificate validation when connecting to a registry. It is supposed to be used for testing purposes only and should not be used in production!
You can set the --snapshotMode=<full (default), time> flag to set how kaniko will snapshot the filesystem.
If --snapshotMode=time is set, only file mtime will be considered when snapshotting (see
limitations related to mtime).
Set this flag to indicate which build stage is the target build stage.
Set this flag as --tarPath=<path> to save the image as a tarball at path instead of pushing the image.
The kaniko executor image is based off of scratch and doesn't contain a shell.
We provide gcr.io/kaniko-project/executor:debug, a debug image which consists of the kaniko executor image along with a busybox shell to enter.
You can launch the debug image with a shell entrypoint:
docker run -it --entrypoint=/busybox/sh gcr.io/kaniko-project/executor:debugkaniko by itself does not make it safe to run untrusted builds inside your cluster, or anywhere else.
kaniko relies on the security features of your container runtime to provide build security.
The minimum permissions kaniko needs inside your container are governed by a few things:
- The permissions required to unpack your base image into it's container
- The permissions required to execute the RUN commands inside the container
If you have a minimal base image (SCRATCH or similar) that doesn't require permissions to unpack, and your Dockerfile doesn't execute any commands as the root user, you can run Kaniko without root permissions. It should be noted that Docker runs as root by default, so you still require (in a sense) privileges to use Kaniko.
You may be able to achieve the same default seccomp profile that Docker uses in your Pod by setting seccomp profiles with annotations on a PodSecurityPolicy to create or update security policies on your cluster.
Similar tools include:
All of these tools build container images with different approaches.
img can perform as a non root user from within a container, but requires that
the img container has RawProc access to create nested containers. kaniko
does not actually create nested containers, so it does not require RawProc
access.
orca-build depends on runc to build images from Dockerfiles, which can not
run inside a container (for similar reasons to img above). kaniko doesn't
use runc so it doesn't require the use of kernel namespacing techniques.
However, orca-build does not require Docker or any privileged daemon (so
builds can be done entirely without privilege).
umoci works without any privileges, and also has no restrictions on the root
filesystem being extracted (though it requires additional handling if your
filesystem is sufficiently complicated). However it has no Dockerfile-like
build tooling (it's a slightly lower-level tool that can be used to build such
builders -- such as orca-build).
buildah requires the same privileges as a Docker daemon does to run, while
kaniko runs without any special privileges or permissions.
FTL and Bazel aim to achieve the fastest possible creation of Docker images
for a subset of images. These can be thought of as a special-case "fast path"
that can be used in conjunction with the support for general Dockerfiles kaniko
provides.
kaniko-users Google group
To Contribute to kaniko, see DEVELOPMENT.md and CONTRIBUTING.md.
When taking a snapshot, kaniko's hashing algorithms include (or in the case of
--snapshotMode=time, only use) a file's
mtime to determine
if the file has changed. Unfortunately there is a delay between when changes to a
file are made and when the mtime is updated. This means:
- With the time-only snapshot mode (
--snapshotMode=time), kaniko may miss changes introduced byRUNcommands entirely. - With the default snapshot mode (
--snapshotMode=full), whether or not kaniko will add a layer in the case where aRUNcommand modifies a file but the contents do not change is theoretically non-deterministic. This does not affect the contents which will still be correct, but it does affect the number of layers.
Note that these issues are currently theoretical only. If you see this issue occur, please open an issue.